ViCorrection: Vietnamese Spelling Correction

Overview

A project to correct spelling errors in Vietnamese texts using Seq2Seq and Seq2Seq-Attention models at character-level

More information can read from training.ipynb or testing.ipynb

Setup

pip install -r requirements.txt

Create dataset

Change .csv path and save_path in dataset/create_dataset.py and run:

cd dataset
python create_dataset.py

Model (character-level)

List of neural models in the repo:

• Seq2Seq
• Seq2Seq-BahdanauAttention

Training

Edit parameters in params.py file and training:

python training.py

Performances

Spell Checker	CER (%)	Full Sequence Acc (%)
Seq2Seq	1.34	82.3
Seq2Seq-Attention	1.12	85.7